search

OpenDriveLab / ST-P3

[ECCV 2022] ST-P3, an end-to-end vision-based autonomous driving framework via spatial-temporal feature learning.
Apache License 2.0
280 stars 34 forks source link

Dear author, questions about depth gt #6

Closed emilyemliyM closed 1 year ago

emilyemliyM commented 1 year ago

I see the depth gt labels is got by the slice operation below:

depth_labels = depths[:, :self.model.receptive_field, :, ::self.model.encoder_downsample, ::self.model.encoder_downsample]

So why is it got by slice operation instead "the encorder"?

Really thanks

charleshsc commented 1 year ago

At the beginning, we first obtain the estimated value of depth by other methods and use it as the gt value of the depth. Note that the depth data obtained at this time is consistent with the size of the original image. (1600 x 900)

Note that the input image goes through the resize process, in order to ensure the consistency the depth data also goes through it, which is " depth_labels = depths[:, :self.model.receptive_field, :, ::self.model.encoder_downsample, ::self.model.encoder_downsample] "

We consider it as the label and no need for further processing of this data except for the resize(slice operation).